এআই এজেন্ট

Chatbot prompt এর জবাব দেয়। কিন্তু AI Agent? সে নিজে decide করে — কোন tool ব্যবহার করবে, কত step এ task complete হবে, কোথায় থামবে। AutoGPT, BabyAGI, Cursor agent, Devin — সব Agent। মানুষ এর মত decision-making AI — NLP এর সর্বোচ্চ frontier।

AI Agent = LLM + Tools + Memory + Planning + Loop। Core pattern: ReAct (Reason → Act → Observe → Repeat)। LLM ভাবে কোন tool call করতে হবে (search, calculator, code, API), tool execute হয়, result observe করে — পরের step decide। Modern frameworks: LangChain Agents, LangGraph, AutoGen, CrewAI। Multi-agent system এ multiple agent collaborate করে।

ভাবুন একজন assistant কে বললেন 'কাল ঢাকা থেকে কক্সবাজার এর flight book কর'। সে: (1) calendar check করে kal এর date, (2) flight search API call, (3) result দেখে cheapest পায়, (4) booking API call, (5) confirmation email পাঠায়। প্রতিটা step এ decide করে পরের কী। AI Agent ঠিক এই — autonomous decision + tool use + multi-step planning।

• Cursor / GitHub Copilot Workspace — coding agent।
• Devin — autonomous SWE agent।
• Perplexity Pro — research agent (multi-search)।
• OpenAI's Operator, Claude Computer Use — browser agent।
• Lovable — full-stack code generation agent।

import os
import json
from openai import OpenAI

client = OpenAI(
    base_url="https://ai.gateway.lovable.dev/v1",
    api_key=os.environ["LOVABLE_API_KEY"],
)

def calculator(expression: str) -> str:
    try:
        return str(eval(expression, {"__builtins__": {}}, {}))
    except Exception as e:
        return f"Error: {e}"

def get_weather(city: str) -> str:
    fake = {"Dhaka": "32C sunny", "London": "15C rainy", "Tokyo": "20C cloudy"}
    return fake.get(city, "Unknown city")

tools = [
    {"type": "function", "function": {
        "name": "calculator",
        "description": "Evaluate a math expression.",
        "parameters": {"type": "object", "properties": {"expression": {"type": "string"}}, "required": ["expression"]},
    }},
    {"type": "function", "function": {
        "name": "get_weather",
        "description": "Get current weather of a city.",
        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]},
    }},
]

def run_agent(user_goal):
    messages = [
        {"role": "system", "content": "You are a helpful agent. Use tools when needed. Reply concisely."},
        {"role": "user", "content": user_goal},
    ]
    for step in range(5):
        res = client.chat.completions.create(
            model="google/gemini-2.5-flash",
            messages=messages,
            tools=tools,
        )
        msg = res.choices[0].message
        messages.append(msg.model_dump(exclude_none=True))

        if not msg.tool_calls:
            print("Final:", msg.content)
            return

        for call in msg.tool_calls:
            args = json.loads(call.function.arguments)
            fn = {"calculator": calculator, "get_weather": get_weather}[call.function.name]
            result = fn(**args)
            print(f"  [tool] {call.function.name}({args}) -> {result}")
            messages.append({"role": "tool", "tool_call_id": call.id, "content": result})

run_agent("What is the weather in Dhaka, and what is 15 percent of 240?")

• Infinite loop — max step limit না দেওয়া।
• Tool description vague — LLM ঠিকমত pick করতে পারে না।
• Error handling না — tool fail হলে agent crash।
• Memory persist না করা — পরের session এ context হারায়।
• Cost track না — agent অনেক LLM call করে।

1. Web search tool যোগ করে real research agent।
2. LangGraph দিয়ে multi-step plan + execute agent।
3. Multi-agent (researcher + writer + critic) workflow।
4. Agent এর action log করে replay তৈরি করুন।

• AI Agent = LLM + Tools + Memory + Loop।
• ReAct pattern = Reason → Act → Observe → Repeat।
• Tool use দিয়ে LLM real world এ action নিতে পারে।
• Frameworks: LangChain, LangGraph, AutoGen, CrewAI।
• Phase 7 complete! আপনি এখন LLM, prompting, RAG, vector DB, agent — modern AI app stack এর master। ChatGPT-level system বানাতে পারবেন।